Something Inc.LoginSchedule a free consultation
GEO

Google's generative AI search control: who should opt out

On August 31 Google finished rolling the generative AI search control out to every Search Console property in the world. It is a one-click switch that removes your site from AI Overviews, AI Mode and Discover's AI features without touching your rankings. Almost nobody should touch it, and the exceptions are specific.

GEODECISION GUIDESEARCH CONSOLE

Google finished the global rollout of two Search Console features on August 31, 2026: the generative AI performance report, and the generative AI search control. The report tells you how often your links show up inside AI answers. The control lets you stop them showing up at all. One of those is useful to every site on the web. The other is a loaded gun with a very short list of legitimate targets.

TL;DR · 60 SECONDSThe generative AI search control is a property-level setting in Search Console that excludes your content from AI Overviews, AI Mode and generative features in Discover. Google states plainly that it is not used as a ranking or inclusion signal anywhere else in Search, so the rankings argument against using it is dead. The real arguments against it are that it is all or nothing until page-level controls arrive in March 2027, that Google will not give you click data to evaluate the tradeoff until December 2026, and that for most B2B and commerce sites the impressions you would be giving up are the top of a funnel you have not finished instrumenting.
Aug 31, 2026
date both features became available to Search Console users worldwide
3
surfaces the control covers: AI Overviews, AI Mode, generative features in Discover
Mar 2027
when page-level controls are due, per the UK regulatory timeline
Dec 2026
when click data for generative AI features is required to appear

This did not arrive because Google wanted to hand publishers a kill switch. The Competition and Markets Authority in the UK made it mandatory in June 2026, implementation began June 17, wider availability followed in July, and the worldwide switch happened at the end of August. Knowing that the control is a regulatory artifact rather than a product decision tells you something useful about how it will evolve: the roadmap dates are compliance deadlines, not feature ambitions.

What the generative AI search control actually does

It sits in Search Console under Settings, on a page called Search generative AI. Flip it, and two things happen to every URL in the property. Your links stop appearing in the covered surfaces, and your content stops being used as an input to generate the answers on those surfaces. That second half is the part teams miss. You are not just declining the citation. You are declining to be part of the source set that shapes what the engine says about your category.

Google puts the propagation window at one to two days for most content, with a few days for the tail, and the setting is reversible whenever you like. Child properties can inherit the parent setting or override it, which is the only granularity available right now and the reason large sites with a single property are stuck with an all or nothing call.

QUESTIONANSWERWHY IT MATTERS
Does it change standard Search rankingsNo. Google states it is not used as a ranking or inclusion signal outside generative AI featuresRemoves the main objection people raise, and shifts the decision onto pure economics
Which surfaces are coveredAI Overviews, AI Mode, and generative AI features in DiscoverThird-party engines like ChatGPT and Perplexity are untouched by this setting
Is it page-levelNo. Property-level only, with inheritance for child propertiesA single property covering docs, blog and product means one decision for all three
Does it stop AI model trainingNo. That is Google-Extended, a separate controlTeams flip this expecting a training opt-out and get neither the outcome nor the visibility
How fast does it applyMost content within one to two days, some longerFast enough to test, slow enough that a panic reversal is not instant
Is it reversibleYes, at any timeThe cost of a wrong call is measured in weeks of lost impressions, not permanence
Does opting out preserve trafficNo. Google states opted-out sites receive no traffic or impressions from generative featuresThere is no partial credit version of this decision
THE ONE SENTENCE VERSIONThe control does exactly one thing well: it removes you from three Google surfaces. Every other outcome people expect from it, less scraping, no model training, leverage in a licensing negotiation, better standard rankings, belongs to a different lever.

The four things the toggle does not do

Most of the bad decisions we expect to see this quarter come from a mismatch between what the switch is for and what someone in a leadership meeting believes it is for. Four misreadings account for nearly all of them.

It is not a training opt-outGoogle-Extended governs whether your content feeds model training. The generative AI search control governs whether your links appear in and inform search-time AI answers. They are separate settings with separate consequences, and flipping the wrong one gets you the cost of both without the benefit of either.
It is not a scraping blockYour content still powers standard Google Search and Google's language understanding systems. Nothing about this setting reduces crawl volume or bandwidth. If the motivation on the table is server load or content theft, this is the wrong instrument entirely.
It is not cross-engineChatGPT, Perplexity, Claude and Copilot do not read this setting. A team that opts out and then reports that the company has withdrawn from AI search has withdrawn from roughly the Google-shaped part of it, and left every other engine exactly as it was.
It is not a negotiating position on its ownWithholding supply is leverage only when the buyer notices the absence. For the overwhelming majority of sites, individually opting out changes nothing about what an AI answer says, because the answer will be assembled from the eleven other sources that cover the same ground.

That fourth point deserves the most weight, because it is the one dressed up as strategy. Collective action by a publishing bloc is leverage. One enterprise site quietly flipping a Search Console setting is a rounding error, and the only party who reliably notices is the team that later has to explain a decline in top-of-funnel impressions.

The decision table

There are real cases for opting out. They are narrower than the discourse implies, and they share a shape: the content is the product, the answer substitutes for the visit, and the site has an existing revenue relationship that does not depend on discovery.

Reference and lookup publisher, ad-funded88%
News publisher with a licensing deal in force71%
B2B software company, considered purchase22%
Ecommerce retailer, transactional intent17%

Illustrative decision weighting for four site profiles, scored on how much of the value is captured before the click. Higher means more of the value is consumed inside the AI answer itself. Illustrative scenario, not measured data.

A reference site whose entire value is a fact someone wanted has already been paid out when the answer renders. A B2B software company whose buyer takes four months and six stakeholders to make a decision has not. The gap between 88 and 22 in that scenario is the whole argument, and it is why the publisher debate and the B2B debate keep talking past each other.

01Opt out only when the answer is the whole productDictionaries, calculators, conversion tables, lyrics, stock quotes, sports scores. If the useful unit of your content fits inside a generated paragraph and your revenue depends on the pageview, exclusion is a coherent commercial position rather than a protest.
02Opt out when a contract requires itSome licensing agreements now carry exclusivity terms that a free appearance in AI Overviews would breach. If legal has told you the content cannot be surfaced without payment, the setting is compliance work, not strategy work, and the analysis is already done.
03Do not opt out to protest measurement gapsThe reporting is genuinely incomplete, and withdrawing from a surface because you cannot measure it is how teams end up with a clean dashboard and a smaller business. Measure the gap instead. We laid out the redundancy approach for exactly this problem in our note on what to do when the analytics stop reporting.
04Do not opt out from a shared propertyIf one Search Console property covers docs, blog, careers and the product marketing site, a property-level exclusion punishes every section to solve a problem in one. Wait for the page-level controls due in March 2027, or split the property first.
THE TEST WE USE WITH CLIENTSAsk what happens to the buyer after the AI answer resolves. If the honest answer is nothing, they got what they came for, exclusion is worth modeling. If the answer is that they now know your category exists and three vendors in it, you are looking at the cheapest brand impression you will ever be offered, and you are considering turning it off.

Why the reporting cannot yet justify an opt-out

The generative AI performance report gives you impressions. It does not give you clicks, and it does not give you position. Under the UK timeline, click data is required by December 2026, which means anyone deciding today is deciding without the single number that would settle it: what an appearance in an AI answer is actually worth in sessions.

There are three more counting quirks worth knowing before you build a case on this data. Multiple results from the same site inside one generative feature count as a single impression, so a page that gets cited three times in one answer looks identical to a page cited once. The report inherits the standard 1,000 row limit, which truncates long tail analysis on any site of size. And the newest days arrive preliminary and marked with dotted lines, so a Monday morning read of the weekend is always an undercount.

METRIC YOU WANTAVAILABLE TODAYWORKAROUND UNTIL DECEMBER
Impressions in AI Overviews and AI ModeYes, by page, country, device and dateNone needed, this is the reliable part
Clicks from generative AI featuresNoSegment landing page sessions against the impression trend and treat the correlation as directional
Position or citation rank inside the answerNoRun a prompt panel on your top commercial questions and log which sources get named
Conversion value of an AI-sourced sessionNoTag and model it downstream, the same way you already handle assisted conversions
Which query produced the impressionNot in the generative reportCross-reference the standard performance report on the same pages and dates

None of that makes the report useless. It makes it a visibility instrument rather than a revenue instrument, which is exactly how we treat citation data generally. The measurement pattern here is the same one we described for tracking citation shifts across model releases: you are watching a trend line for shape, not reading a P and L.

Withdrawing from a surface because you cannot measure it is not caution. It is choosing a smaller business with a tidier dashboard.

There is also a real placement question underneath this. NewzDash found that nearly one in six US trending news queries, 15.5 percent, now place Top Stories inside AI Overviews, with the UK figure at 17.46 percent. For a news operation, opting out does not just remove an AI citation. It can remove a carousel placement that used to sit in a straightforward organic slot. Rahul Jain of Noble put the publisher case bluntly: opting out of AI Overviews will hurt most publishers more than it helps. Search Engine Journal collected those figures and reactions as the control reached general availability.

What to do instead of flipping the switch

For roughly 95 percent of the sites reading this, the correct action this month is to open the report, not the control. Establish the baseline while the data is young, because the first thing you will want in March 2027, when page-level controls arrive and the decision gets genuinely interesting, is six months of history showing which sections earn AI impressions and which do not.

Then work the supply side. A page that is easy to quote correctly earns citations across every engine at once, and the structural work that wins an AI Overview is the same work that wins a Perplexity citation. That starts with machine-readable structure, which we set out in the enterprise standard for structured data in AI search, and it compounds in a way that a defensive setting never will. Our generative engine optimization work is built almost entirely on that premise, and the pattern held across the Zenity engagement, where the same structural changes moved both classic rankings and citations in four engines.

One more piece of housekeeping while you are in Settings. Check what Google-Extended is set to, because that is the control most teams believe they have already configured and most have not. The two settings answer different questions and a company can reasonably want opposite answers to them: appear in the search-time answers, decline to feed the training corpus. You cannot get that combination by guessing which toggle is which.

DO THIS NEXTOpen Search Console, confirm the generative AI performance report is populating for every property you own, and export the last full month by page before the data ages out of easy reach. Then check the Search generative AI control setting on each property and write down what it is currently set to, because on a large account with several owners, at least one property is usually set to something nobody chose deliberately. Only after those two things are documented is the opt-out question worth an actual meeting.

See where you are cited today

A free snapshot audit of your rankings and AI citations before we ever talk.

JB
Josh BernsteinMANAGING PARTNER, SOMETHING INC.

Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.

Free consultation

Let us be the last SEO agency you ever work with

A 30 minute call and a free audit of your SEO and GEO position. You keep the findings either way.